agent-desktop lets coding agents drive desktop apps through the accessibility tree, not screenshots
Coding agents can operate desktop apps by the names of their buttons and fields instead of guessing from screenshots. If the screen changed, a click fails instead of hitting the wrong thing. Snapshots can capture private desktop content. Why it matters: Screenshot-based computer use guesses where to click from pixels. Reading the app's accessibility tree gives the agent named buttons and fields, and a stale reference fails loudly instead of hitting the wrong row. The limit is the same tree: in our reading, apps with poor accessibility labels give the agent little to work with, and the README warns that snapshots and clipboard reads can contain private desktop content.
Try this: Install it with `npm install -g agent-desktop`, run `agent-desktop snapshot --app Finder -i` (or any app from `list-apps`), and check whether the refs name the buttons your agent would need.